mongo 2.15.0 → 2.17.0

data/lib/mongo/cluster.rb

@@ -159,6 +159,10 @@ module Mongo
       @sdam_flow_lock = Mutex.new
       Session::SessionPool.create(self)
 
+      if seeds.empty? && load_balanced?
+        raise ArgumentError, 'Load-balanced clusters with no seeds are prohibited'
+      end
+
       # The opening topology is always unknown with no servers.
       # https://github.com/mongodb/specifications/pull/388
       opening_topology = Topology::Unknown.new(options, monitoring, self)
@@ -171,7 +175,7 @@ module Mongo
       @seeds = seeds = seeds.uniq
       servers = seeds.map do |seed|
         # Server opening events must be sent after topology change events.
-        # Therefore separate server addition, done here before topoolgy change
+        # Therefore separate server addition, done here before topology change
         # event is published, from starting to monitor the server which is
         # done later.
         add(seed, monitor: false)
@@ -179,11 +183,20 @@ module Mongo
 
       if seeds.size >= 1
         # Recreate the topology to get the current server list into it
-        @topology = topology.class.new(topology.options, topology.monitoring, self)
-        publish_sdam_event(
-          Monitoring::TOPOLOGY_CHANGED,
-          Monitoring::Event::TopologyChanged.new(opening_topology, @topology)
-        )
+        recreate_topology(topology, opening_topology)
+      end
+
+      if load_balanced?
+        # We are required by the specifications to produce certain SDAM events
+        # when in load-balanced topology.
+        # These events don't make a lot of sense from the standpoint of the
+        # driver's SDAM implementation, nor from the standpoint of the
+        # driver's load balancer implementation.
+        # They are just required boilerplate.
+        #
+        # Note that this call must be done above the monitoring_io check
+        # because that short-circuits the rest of the constructor.
+        fabricate_lb_sdam_events_and_set_server_type
       end
 
       if options[:monitoring_io] == false
@@ -203,7 +216,7 @@ module Mongo
       @connected = true
 
       if options[:cleanup] != false
-        @cursor_reaper = CursorReaper.new
+        @cursor_reaper = CursorReaper.new(self)
         @socket_reaper = SocketReaper.new(self)
         @periodic_executor = PeriodicExecutor.new([
           @cursor_reaper, @socket_reaper,
@@ -212,54 +225,56 @@ module Mongo
         @periodic_executor.run!
       end
 
-      # Need to record start time prior to starting monitoring
-      start_monotime = Utils.monotonic_time
-
-      servers.each do |server|
-        server.start_monitoring
-      end
+      unless load_balanced?
+        # Need to record start time prior to starting monitoring
+        start_monotime = Utils.monotonic_time
 
-      if options[:scan] != false
-        server_selection_timeout = options[:server_selection_timeout] || ServerSelector::SERVER_SELECTION_TIMEOUT
-        # The server selection timeout can be very short especially in
-        # tests, when the client waits for a synchronous scan before
-        # starting server selection. Limiting the scan to server selection time
-        # then aborts the scan before it can process even local servers.
-        # Therefore, allow at least 3 seconds for the scan here.
-        if server_selection_timeout < 3
-          server_selection_timeout = 3
+        servers.each do |server|
+          server.start_monitoring
         end
-        deadline = start_monotime + server_selection_timeout
-        # Wait for the first scan of each server to complete, for
-        # backwards compatibility.
-        # If any servers are discovered during this SDAM round we are going to
-        # wait for these servers to also be queried, and so on, up to the
-        # server selection timeout or the 3 second minimum.
-        loop do
-          # Ensure we do not try to read the servers list while SDAM is running
-          servers = @sdam_flow_lock.synchronize do
-            servers_list.dup
-          end
-          if servers.all? { |server| server.last_scan_monotime && server.last_scan_monotime >= start_monotime }
-            break
-          end
-          if (time_remaining = deadline - Utils.monotonic_time) <= 0
-            break
+
+        if options[:scan] != false
+          server_selection_timeout = options[:server_selection_timeout] || ServerSelector::SERVER_SELECTION_TIMEOUT
+          # The server selection timeout can be very short especially in
+          # tests, when the client waits for a synchronous scan before
+          # starting server selection. Limiting the scan to server selection time
+          # then aborts the scan before it can process even local servers.
+          # Therefore, allow at least 3 seconds for the scan here.
+          if server_selection_timeout < 3
+            server_selection_timeout = 3
           end
-          log_debug("Waiting for up to #{'%.2f' % time_remaining} seconds for servers to be scanned: #{summary}")
-          # Since the semaphore may have been signaled between us checking
-          # the servers list above and the wait call below, we should not
-          # wait for the full remaining time - wait for up to 1 second, then
-          # recheck the state.
-          begin
-            server_selection_semaphore.wait([time_remaining, 1].min)
-          rescue ::Timeout::Error
-            # nothing
+          deadline = start_monotime + server_selection_timeout
+          # Wait for the first scan of each server to complete, for
+          # backwards compatibility.
+          # If any servers are discovered during this SDAM round we are going to
+          # wait for these servers to also be queried, and so on, up to the
+          # server selection timeout or the 3 second minimum.
+          loop do
+            # Ensure we do not try to read the servers list while SDAM is running
+            servers = @sdam_flow_lock.synchronize do
+              servers_list.dup
+            end
+            if servers.all? { |server| server.last_scan_monotime && server.last_scan_monotime >= start_monotime }
+              break
+            end
+            if (time_remaining = deadline - Utils.monotonic_time) <= 0
+              break
+            end
+            log_debug("Waiting for up to #{'%.2f' % time_remaining} seconds for servers to be scanned: #{summary}")
+            # Since the semaphore may have been signaled between us checking
+            # the servers list above and the wait call below, we should not
+            # wait for the full remaining time - wait for up to 1 second, then
+            # recheck the state.
+            begin
+              server_selection_semaphore.wait([time_remaining, 1].min)
+            rescue ::Timeout::Error
+              # nothing
+            end
           end
         end
-      end
 
-      start_stop_srv_monitor
+        start_stop_srv_monitor
+      end
     end
 
     # Create a cluster for the provided client, for use when we don't want the
@@ -321,6 +336,14 @@ module Mongo
     def_delegators :topology, :replica_set?, :replica_set_name, :sharded?,
                    :single?, :unknown?
 
+    # Returns whether the cluster is configured to be in the load-balanced
+    # topology.
+    #
+    # @return [ true | false ] Whether the topology is load-balanced.
+    def load_balanced?
+      topology.is_a?(Topology::LoadBalanced)
+    end
+
     [:register_cursor, :schedule_kill_cursor, :unregister_cursor].each do |m|
       define_method(m) do |*args|
         if options[:cleanup] != false
@@ -599,9 +622,25 @@ module Mongo
     #   on 4.2+ servers).
     # @option aptions [ true | false ] :awaited Whether the updated description
     #   was a result of processing an awaited hello.
+    # @option options [ Object ] :service_id Change state for the specified
+    #   service id only.
     #
     # @api private
     def run_sdam_flow(previous_desc, updated_desc, options = {})
+      if load_balanced?
+        if updated_desc.config.empty?
+          unless options[:keep_connection_pool]
+            servers_list.each do |server|
+              # TODO should service id be taken out of updated_desc?
+              # We could also assert that
+              # options[:service_id] == updated_desc.service_id
+              server.clear_connection_pool(service_id: options[:service_id])
+            end
+          end
+        end
+        return
+      end
+
       @sdam_flow_lock.synchronize do
         flow = SdamFlow.new(self, previous_desc, updated_desc,
           awaited: options[:awaited])
@@ -782,8 +821,15 @@ module Mongo
     def add(host, add_options=nil)
       address = Address.new(host, options)
       if !addresses.include?(address)
-        server = Server.new(address, self, @monitoring, event_listeners, options.merge(
-          monitor: false))
+        opts = options.merge(monitor: false)
+        # Note that in a load-balanced topology, every server must be a
+        # load balancer (load_balancer: true is specified in the options)
+        # but this option isn't set here because we are required by the
+        # specifications to pretent the server started out as an unknown one
+        # and publish server description change event into the load balancer
+        # one. The actual correct description for this server will be set
+        # by the fabricate_lb_sdam_events_and_set_server_type method.
+        server = Server.new(address, self, @monitoring, event_listeners, opts)
         @update_lock.synchronize do
           # Need to recheck whether server is present in @servers, because
           # the previous check was not under a lock.
@@ -899,6 +945,10 @@ module Mongo
     #
     # @api private
     def validate_session_support!
+      if topology.is_a?(Topology::LoadBalanced)
+        return
+      end
+
       @state_change_lock.synchronize do
         @sdam_flow_lock.synchronize do
           if topology.data_bearing_servers?
@@ -982,6 +1032,41 @@ module Mongo
       msg = "The deployment that the driver is connected to does not support sessions: #{reason}"
       raise Error::SessionsNotSupported, msg
     end
+
+    def fabricate_lb_sdam_events_and_set_server_type
+      # Although there is no monitoring connection in load balanced mode,
+      # we must emit the following series of SDAM events.
+      server = @servers.first
+      # We are guaranteed to have the server here.
+      server.publish_opening_event
+      server_desc = server.description
+      # This is where a load balancer actually gets its correct server
+      # description.
+      server.update_description(
+        Server::Description.new(server.address, {},
+          load_balancer: true,
+          force_load_balancer: options[:connect] == :load_balanced,
+        )
+      )
+      publish_sdam_event(
+        Monitoring::SERVER_DESCRIPTION_CHANGED,
+        Monitoring::Event::ServerDescriptionChanged.new(
+          server.address,
+          topology,
+          server_desc,
+          server.description
+        )
+      )
+      recreate_topology(topology, topology)
+    end
+
+    def recreate_topology(new_topology_template, previous_topology)
+      @topology = topology.class.new(new_topology_template.options, new_topology_template.monitoring, self)
+      publish_sdam_event(
+        Monitoring::TOPOLOGY_CHANGED,
+        Monitoring::Event::TopologyChanged.new(previous_topology, @topology)
+      )
+    end
   end
 end
 

data/lib/mongo/collection/view/aggregation.rb

@@ -73,6 +73,25 @@ module Mongo
         # @param [ Array<Hash> ] pipeline The pipeline of operations.
         # @param [ Hash ] options The aggregation options.
         #
+        # @option options [ true, false ] :allow_disk_use Set to true if disk
+        #   usage is allowed during the aggregation.
+        # @option options [ Integer ] :batch_size The number of documents to return
+        #   per batch.
+        # @option options [ true, false ] :bypass_document_validation Whether or
+        #   not to skip document level validation.
+        # @option options [ Hash ] :collation The collation to use.
+        # @option options [ String ] :comment Associate a comment with the aggregation.
+        # @option options [ String ] :hint The index to use for the aggregation.
+        # @option options [ Hash ] :let Mapping of variables to use in the pipeline.
+        #   See the server documentation for details.
+        # @option options [ Integer ] :max_time_ms The maximum amount of time in
+        #   milliseconds to allow the aggregation to run.
+        # @option options [ true, false ] :use_cursor Indicates whether the command
+        #   will request that the server provide results using a cursor. Note that
+        #   as of server version 3.6, aggregations always provide results using a
+        #   cursor and this option is therefore not valid.
+        # @option options [ Session ] :session The session to use.
+        #
         # @since 2.0.0
         def initialize(view, pipeline, options = {})
           @view = view
@@ -108,42 +127,63 @@ module Mongo
           @view.send(:server_selector)
         end
 
-        def aggregate_spec(session)
-          Builder::Aggregation.new(pipeline, view, options.merge(session: session)).specification
+        def aggregate_spec(server, session, read_preference)
+          Builder::Aggregation.new(
+            pipeline,
+            view,
+            options.merge(session: session, read_preference: read_preference)
+          ).specification
         end
 
         def new(options)
           Aggregation.new(view, pipeline, options)
         end
 
-        def initial_query_op(session)
-          Operation::Aggregate.new(aggregate_spec(session))
+        def initial_query_op(server, session, read_preference)
+          Operation::Aggregate.new(aggregate_spec(server, session, read_preference))
         end
 
-        def valid_server?(server)
-          description = server.with_connection do |connection|
-            connection.description
+        # Return effective read preference for the operation.
+        #
+        # If the pipeline contains $merge or $out, and read preference specified
+        # by user is secondary or secondary_preferred, and selected server is below
+        # 5.0, than this method returns primary read preference, because the
+        # aggregation will be routed to primary. Otherwise return the original
+        # read preference.
+        #
+        # See https://github.com/mongodb/specifications/blob/master/source/crud/crud.rst#read-preferences-and-server-selection
+        #
+        # @param [ Server ] server The server on which the operation
+        #   should be executed.
+        # @return [ Hash | nil ] read preference hash that should be sent with
+        #   this command.
+        def effective_read_preference(server)
+          return unless view.read_preference
+          return view.read_preference unless write?
+          return view.read_preference unless [:secondary, :secondary_preferred].include?(view.read_preference[:mode])
+
+          primary_read_preference = {mode: :primary}
+          if server.primary?
+            log_warn("Routing the Aggregation operation to the primary server")
+            primary_read_preference
+          elsif server.mongos? && !server.features.merge_out_on_secondary_enabled?
+            log_warn("Routing the Aggregation operation to the primary server")
+            primary_read_preference
+          else
+            view.read_preference
           end
-          description.standalone? || description.mongos? || description.primary? || secondary_ok?
-        end
 
-        def secondary_ok?
-          !write?
         end
 
         def send_initial_query(server, session)
-          unless valid_server?(server)
-            log_warn("Rerouting the Aggregation operation to the primary server - #{server.summary} is not suitable")
-            server = cluster.next_primary(nil, session)
-          end
-          validate_collation!(server)
-          initial_query_op(session).execute(server, context: Operation::Context.new(client: client, session: session))
-        end
-
-        def validate_collation!(server)
-          if options[:collation] && !server.with_connection { |connection| connection.features }.collation_enabled?
-            raise Error::UnsupportedCollation.new
-          end
+          initial_query_op(
+            server,
+            session,
+            effective_read_preference(server)
+          ).execute(
+              server,
+              context: Operation::Context.new(client: client, session: session)
+            )
         end
 
         # Skip, sort, limit, projection are specified as pipeline stages

data/lib/mongo/collection/view/builder/aggregation.rb

@@ -30,16 +30,17 @@ module Mongo
           #
           # @since 2.2.0
           MAPPINGS = BSON::Document.new(
-            :allow_disk_use => 'allowDiskUse',
-            :max_time_ms => 'maxTimeMS',
+            allow_disk_use: 'allowDiskUse',
+            bypass_document_validation: 'bypassDocumentValidation',
+            explain: 'explain',
+            collation: 'collation',
+            comment: 'comment',
+            hint: 'hint',
+            let: 'let',
             # This is intentional; max_await_time_ms is an alias for maxTimeMS
             # used on getMore commands for change streams.
-            :max_await_time_ms => 'maxTimeMS',
-            :explain => 'explain',
-            :bypass_document_validation => 'bypassDocumentValidation',
-            :collation => 'collation',
-            :hint => 'hint',
-            :comment => 'comment'
+            max_await_time_ms: 'maxTimeMS',
+            max_time_ms: 'maxTimeMS',
           ).freeze
 
           def_delegators :@view, :collection, :database, :read, :write_concern
@@ -55,12 +56,9 @@ module Mongo
 
           # Initialize the builder.
           #
-          # @example Initialize the builder.
-          #   Aggregation.new(map, reduce, view, options)
-          #
           # @param [ Array<Hash> ] pipeline The aggregation pipeline.
           # @param [ Collection::View ] view The collection view.
-          # @param [ Hash ] options The map/reduce options.
+          # @param [ Hash ] options The map/reduce and read preference options.
           #
           # @since 2.2.0
           def initialize(pipeline, view, options)
@@ -79,11 +77,12 @@ module Mongo
           # @since 2.2.0
           def specification
             spec = {
-                    selector: aggregation_command,
-                    db_name: database.name,
-                    read: view.read_preference,
-                    session: @options[:session]
-                   }
+              selector: aggregation_command,
+              db_name: database.name,
+              read: @options[:read_preference] || view.read_preference,
+              session: @options[:session],
+              collation: @options[:collation],
+            }
             if write?
               spec.update(write_concern: write_concern)
             end

data/lib/mongo/collection/view/builder/map_reduce.rb

@@ -36,7 +36,7 @@ module Mongo
             scope: 'scope',
             verbose: 'verbose',
             bypass_document_validation: 'bypassDocumentValidation',
-            collation: 'collation'
+            collation: 'collation',
           ).freeze
 
           def_delegators :@view, :collection, :database, :filter, :read, :write_concern
@@ -71,36 +71,6 @@ module Mongo
             @options = options
           end
 
-          # Get the specification for issuing a find command on the map/reduce
-          # results.
-          #
-          # @example Get the command specification.
-          #   builder.command_specification
-          #
-          # @return [ Hash ] The specification.
-          #
-          # @since 2.2.0
-          def command_specification
-            {
-              selector: find_command,
-              db_name: query_database,
-              read: read,
-              session: options[:session]
-            }
-          end
-
-          # Get the specification for the document query after a map/reduce.
-          #
-          # @example Get the query specification.
-          #   builder.query_specification
-          #
-          # @return [ Hash ] The specification.
-          #
-          # @since 2.2.0
-          def query_specification
-            { selector: {}, options: {}, db_name: query_database, coll_name: query_collection }
-          end
-
           # Get the specification to pass to the map/reduce operation.
           #
           # @example Get the specification.
@@ -113,6 +83,8 @@ module Mongo
             spec = {
               selector: map_reduce_command,
               db_name: database.name,
+              # Note that selector just above may also have a read preference
+              # specified, per the #map_reduce_command method below.
               read: read,
               session: options[:session]
             }
@@ -121,8 +93,6 @@ module Mongo
 
           private
 
-          OUT_ACTIONS = [ :replace, :merge, :reduce ].freeze
-
           def write?(spec)
             if out = spec[:selector][:out]
               out.is_a?(String) ||
@@ -130,37 +100,30 @@ module Mongo
             end
           end
 
-          def find_command
-            BSON::Document.new('find' => query_collection, 'filter' => {})
-          end
-
           def map_reduce_command
             command = BSON::Document.new(
               :mapReduce => collection.name,
               :map => map,
               :reduce => reduce,
               :query => filter,
-              :out => { inline: 1 }
+              :out => { inline: 1 },
             )
+            # Shouldn't this use self.read ?
             if collection.read_concern
               command[:readConcern] = Options::Mapper.transform_values_to_strings(
                 collection.read_concern)
             end
-            command.merge!(view_options)
+            command.update(view_options)
+            command.update(options.slice(:collation))
+            # Read preference isn't simply passed in the command payload
+            # (it may need to be converted to wire protocol flags)
+            # so remove it here and hopefully it's handled elsewhere.
+            # If not, RUBY-2706.
+            command.delete(:read)
             command.merge!(Options::Mapper.transform_documents(options, MAPPINGS))
             command
           end
 
-          def query_database
-            options[:out].respond_to?(:keys) && options[:out][:db] ? options[:out][:db] : database.name
-          end
-
-          def query_collection
-            if options[:out].respond_to?(:keys)
-              options[:out][OUT_ACTIONS.find { |action| options[:out][action] }]
-            end || options[:out]
-          end
-
           def view_options
             @view_options ||= (opts = view.options.dup
                                opts.delete(:session)

data/lib/mongo/collection/view/builder.rb

@@ -17,7 +17,3 @@
 
 require 'mongo/collection/view/builder/aggregation'
 require 'mongo/collection/view/builder/map_reduce'
-require 'mongo/collection/view/builder/op_query'
-require 'mongo/collection/view/builder/find_command'
-require 'mongo/collection/view/builder/flags'
-require 'mongo/collection/view/builder/modifiers'

data/lib/mongo/collection/view/change_stream.rb

@@ -298,8 +298,8 @@ module Mongo
           [{ '$changeStream' => change_doc }] + @change_stream_filters
         end
 
-        def aggregate_spec(session)
-          super(session).tap do |spec|
+        def aggregate_spec(server, session, read_preference)
+          super(server, session, read_preference).tap do |spec|
             spec[:selector][:aggregate] = 1 unless for_collection?
           end
         end
@@ -349,7 +349,11 @@ module Mongo
         end
 
         def send_initial_query(server, session)
-          initial_query_op(session).execute(server, context: Operation::Context.new(client: client, session: session))
+          initial_query_op(server, session, view.read_preference)
+            .execute(
+              server,
+              context: Operation::Context.new(client: client, session: session)
+            )
         end
 
         def time_to_bson_timestamp(time)